Animesh Garg describes robotics as being early in its frontier AI transition. He divides generalizable autonomy into three linked problems: representing the environment, building world or reward models that guide action, and using reinforcement learning or continual improvement to refine behavior.
Animesh Garg argues that affordable, standardized developer hardware can accelerate experimentation in the same way early personal computers did. He points to low-cost small robots and developer-focused humanoids as platforms that let more researchers collect data, build simulators and test applications.
Animesh Garg predicts that useful humanoids could fall into a $10,000 to $15,000 range within two years, while stressing that robot data remains the agreed scaling ingredient even as model architectures evolve. He expects vision-language-action models, world models and real-to-simulation approaches to converge, with spatial memory and three-dimensional context remaining major unsolved challenges.
Watch on YouTube
